Thomas Metzinger (2007; revised to 12 November 2013), “Self models”, Scholarpedia 2(10):4174 — DOI 10.4249/scholarpedia.4174
Key takeaways
- SMT. The self-model theory of subjectivity is both a philosophical theory of the phenomenal self and the first-person perspective and an interdisciplinary research programme, worked at phenomenological, representational, functional and neural levels.
- No self, a self-model. “There is no such thing as a substantial self … but only a dynamic, ongoing process creating very specific representational and functional properties.” The phenomenal self-model (PSM) is “that partition of an organism’s self-model which is conscious” because it meets further functional constraints, chiefly integration into a “virtual window of presence” and a single world model.
- Transparency is the key. A representation is transparent when its earlier processing stages are unavailable to attention, so one “sees through” it to its content. “The central claim of SMT is that a phenomenal self emerges if and only if a system operates under a transparent PSM.” The prediction: a fully opaque PSM would end the experience of selfhood.
- The first-person perspective is a transparent phenomenal model of the intentionality relation (PMIR): the system models itself as directed at objects and goals and does not recognise the model as a model.
- Self-models can be unconscious and artificial. Bongard, Zykov and Lipson’s “starfish” robot infers its own body from actuation–sensation data, chooses actions that discriminate between 15 candidate self-models, and after losing part of a leg revises its model and “learns to limp”. “You do not have to be a living being in order to have a self-model.”
- Three orders of embodiment. First-order: intelligence from body and environment without explicit computation. Second-order: an explicit, coherent self-representation as an embodied agent, used for control (the starfish, many animals). Third-order: some of that content is conscious, a PSM (waking humans, perhaps orangutans).
Standing in the record
Filed on 2026-10-02 at the owner’s request for Metzinger’s self-model theory, as the lawfully readable statement of it, beside the book (LIT-467, Deferred), and to relate it to Minsky’s self-models in The Emotion Machine (LIT-432). No anthology topic holds it. It carries no instruction for machine-learning practice; its machine example is a robotics result reported as illustration. The anthology does not hold it.
NOTE-372 is the reading, and it placed the work: Active. It is short and by the theory’s author, and states the central claim exactly. It argues little: the transparency claim is a definition with a prediction, and the empirical examples are cited. Its online text has drifted from the accepted version, so cite it by revision date.
Where it meets the record, set out in the NOTE:
- Minsky (LIT-432, NOTE-348). Both deny a substantial self and make the self a matter of self-models. Minsky’s Self is “a network of models” switched among, and a single unified model would be impractical. Metzinger’s PSM is “a coherent self-representation, a consistent internal model of itself as a whole”, an instrument of global control. They differ on one model or many, and on what produces the sense of a self: Minsky’s convenient Single-Self idea, Metzinger’s transparency.
- Dennett (LIT-442). Both are no-self views. Dennett’s self is a posit of interpretation, the brain included as interpreter; Metzinger’s is the content of a model the brain runs and cannot see as a model.
- Machines and consciousness (THEORY-023). The article separates having a self-model (a robot can) from having a phenomenal self (needs further constraints, “presently still unknown”), so a self-model is not evidence of experience on its own.